Directory Structure:

AgenticContentCreator/

├── agents/
│   ├── content_generator.py
│   ├── feedback_manager.py
│   └── rag_retriever.py
├── publisher/
│   └── publisher.py
├── storage/
│   └── aws_s3_manager.py
├── agentic_pipeline/
│   └── autogen_content_pipeline.py  
├── main.py
├── requirements.txt

Agentic AI Content Generator using free LLM and Stable Diffusion.

In this project, I have built an Agentic AI-powered content creation ecosystem—essentially a self-improving system that automates generating, refining, and publishing social media content using multiple AI components. Here’s what’s happening step-by-step:

Content Generation:

Microsoft Autogen + LangChain + LLMs + DALL·E + RAG: These tools generate rich content (text + images). Autogen and LangChain coordinate different models (like LLMs for text, DALL·E for images) to create end-to-end posts.

Content Publishing:

The system publishes this AI-generated content automatically across multiple social media platforms.

Feedback Loop with MemGPT:

After publishing, MemGPT captures user interaction data (likes, comments, shares, engagement metrics).

This data is used to refine the content generation pipeline, effectively making your AI learn what content performs better and adapt future outputs.

RAG (Retrieval-Augmented Generation):

You enrich content by pulling relevant, real-time information from external databases or APIs before generation, ensuring content is more contextually relevant or timely.

TensorFlow + AWS Deployment:

TensorFlow may be used for specific model serving or analytics tasks.

AWS handles deployment, scalability, and reliability, ensuring that your system can operate and scale in production environments.

In short, you’re automating the full content pipeline: generate → publish → analyze → refine, using multiple modern AI tools working together in an "agentic" (goal-driven, self-directed) manner.

• Content Generation: OpenAI LLM API + DALL·E API

• Feedback Loop: Simulated with MemGPT-like memory via local storage

• RAG Enrichment: Simple document retrieval using FAISS

• Publishing: Simulated (you can replace it with real API calls)

• Deployment: AWS S3 for storing content, Lambda trigger simulation

Components:

• OpenAI GPT 5 Text Generation API
• DALLE
• AUTOGEN
• LangChain
• MemGPT
• AWS

Virtual Environment Creation

python -m venv .venv

To Activate the Virtual Environment

source .venv/bin/activate

To Deactivate the Virtual Environment

deactivate 

Run Instructions:

pip freeze > requirements.txt

Set Reddit API ACCESS TOKEN

• Go to https://www.reddit.com/settings/privacy, click on Third-party app authorizations under Advanced
• Under apps, click on 'create another app'
• Select script
• Add the following information - Name: AgenticAI Publisher, Type: Script, Redirect URI: http://localhost:8080
• Save Client ID and Secret

Set Environmental Variables

Stores the environmental variables in the .env file in the same format.  

export OPENAI_API_KEY="your-openai-api-key"

export REDDIT_API_KEY = "your-reddit-api-key"

aws configure

Run the simplified version

python main.py

Run AutoGen Multi-Agent Pipeline

cd agentic_pipeline

python autogen_content_pipeline.py